1. Evolutionary Programming An Example Evolutionary Computation Procedure EC{ t = 0; Initialize P(t); Evaluate P(t); While (Not Done) { Parents(t) = Select_Parents(P(t)); Offspring(t) = Procreate(Parents(t)); Evaluate(Offspring(t)); P(t+1)= Select_Survivors(P(t),Offspring(t)); t = t + 1; }

2. Evolutionary Programming In EC, the replacement method could also be called Select_Survivors. Therefore, during this course, we will discuss two forms of selection: –Parent Selection (Select_Parents(Pop(t))), and –Survivor Selection (Select_Survivors(Pop(t), Offspring(t)). In EC, there are basically two types of survivor selection methods: –(  + ) and –( , )

3. Evolutionary Programming In (  + ) selection, –The population of  individuals (where  = |P(t)|, the population size) are used to create offspring (where = |Offspring(t)|). –The best  of P(t)  Offspring(t) are selected to survive. How might a (  +1)-EC operate?

4. Evolutionary Programming In ( , ) selection, –The population of  individuals (where  = |P(t)|, the population size) are used to create offspring (where = |Offspring(t)|). –The best  of Offspring(t) are selected to survive. What constraints must be placed on by a ( , )-EC?

5. Evolutionary Programming: Early EP Early evolutionary programs evolved populations of finite state machines in an effort to solve prediction problems [see http://www.natural-selection.com/Library/1998/Revisiting_EP.pdf ]. http://www.natural-selection.com/Library/1998/Revisiting_EP.pdf Early EP used 5 mutation operators: –Add a State, –Delete a State, –Mutate the Start State, –Mutate a Link, and –Mutate an output symbol.

6. Evolutionary Programming: Early EP The fitness of an individual machine was mean absolute error in predicting an output symbol given an input symbol. Each of the  machines were allowed to create one offspring by applying 1-5 of the mutations presented earlier. The population of  offspring machines were exposed to an environment, in the form of a string of s symbols.

7. Evolutionary Programming: Early EP The best  of the 2  machines were selected to survive. –For  > 9, q-tournament selection was applied to develop a subjective fitness. In this form of selection: Each individual competes with q (typically less than 10) randomly selected machines, The subjective fitness for a machine in the number of individuals (of the q randomly selected individuals) that it is better than. The process of procreation, evaluation, and selection was repeated for a total of 5 generations.

8. Evolutionary Programming: Early EP After every 5 generations the best machine was allowed to predict the (s+1) symbol. After the prediction, the true (s+1) was added to the environment. This process was repeated for M-1 cycles (with the final length of the environment being (s+M)).

9. Evolutionary Programming: Standard EP David B. Fogel adapted EP for solving parameter optimization problems in the late 1980’s (Bäck 1996). Standard EP, like its predecessor, attempts to solve minimization problems. Let’s take a look at it!

10. Evolutionary Programming: Standard EP Procedure standardEP{ t = 0; Initialize P(t); /* of  individuals */ Evaluate P(t); while (t <= (4000-  )/  ){ for (i=0; i<  ; i++){ Create_Offspring(, ): x  +i = x i + sqrt(fit i )N x (0,1); y  +i = y i + sqrt(fit i )N y (0,1); fit  +i = Evaluate( ); } Compute Subjective Fitness if   10; P(t+1) = Best  of the 2  individuals; t = t + 1; }

11. Evolutionary Programming: Continuous Standard EP Procedure Continuous_standardEP{ t = 0; Initialize P(t); /* of  individuals */ Evaluate P(t); while (t <= (4000-  )){ Create_Offspring(, ): x  = x i + sqrt(fit i )N x (0,1); y  = y i + sqrt(fit i )N y (0,1); fit  = Evaluate( ); P(t+1) = Best  of the  +1 individuals; t = t + 1; }

12. Evolutionary Programming: Meta-EP Procedure metaEP{ t = 0; Initialize P(t); /* of  individuals */ Evaluate P(t); while (t <= (4000-  )/  ){ for (i=0; i<  ; i++){ Create_Offspring(, ): x  +i = x i +  i,x N x (0,1);   +i,x =  i,x +   i,x N x (0,1); y  +i = y i +  i,y N y (0,1);   +i,y =  i,y +   i,y N y (0,1); fit  +i = Evaluate( ); } Compute Subjective Fitness if   10; P(t+1) = Best  of the 2  individuals; t = t + 1; }

13. Evolutionary Programming: Continuous Meta-EP Procedure continuous_metaEP{ t = 0; Initialize P(t); /* of  individuals */ Evaluate P(t); while (t <= (4000-  )){ Create_Offspring(, ): x  = x i +  i,x N x (0,1);  ,x =  i,x +   i,x N x (0,1); y  = y i +  i,y N y (0,1);  ,y =  i,y +   i,y N y (0,1); fit  = Evaluate( ); P(t+1) = Best  of the  +1 individuals; t = t + 1; }